Representable correcting terms for possibly underflowing floating point operations | IEEE Conference Publication | IEEE Xplore